A Hundred Dollars of Future

I had a feature I was never going to build.

Semantic search over a fully automated image processing and multimodal embedding pipeline — the kind of thing that sounds impressive in a pitch deck and lives forever on a dream list because you keep finding better excuses not to start it. I'd allocated two to three days for the backend alone, with "unknown" penciled in for the frontend, which is engineer for "I'll figure that out when I'm forced to."

Claude Code finished both passes in ten minutes.

I'm not being cute about the timeline. I mean I blinked and there was a working implementation, properly integrated into the gas meter and user usage tracking, ready to show why a customer needs the thing before they've even asked for it. The kind of feature that previously required someone to beg loudly and repeatedly — now it just exists, because I had a Thursday afternoon and a hundred dollars.

That's the number that keeps rattling around. Just under a hundred dollars for an entire dream feature list, crushed in twenty-four hours, the kind of velocity that makes you recalculate every past decision about what was or wasn't worth building.

The thing thinks like a very smart person and works like one — which sounds obvious until you've used every other coding assistant and noticed they all think like a very fast autocomplete. The difference isn't speed. It's that it understands what you're actually trying to do and then goes and does it, and then comes back and tells you it also noticed you'd want it wired into the usage tracking, so it handled that too.

The future of good software is just compute cycles you can afford to burn. That's it. That's the whole insight. The limiting factor stopped being skill or time and became whether you're willing to pay the bill — which is either the most democratizing thing that's ever happened to software development or the most clarifying indictment of how much of what we called "engineering work" was really just friction wearing a hard hat.

Anthropic is going to print money if they stay this far ahead.

I'm going to go add the features I put off last year.